Labeling examples is usually an expensive activity in the real world. The focus of this presentation is methods that attempt to reduce the labeling cost assosiated with building models when using supervsed learning methods. To this end we will discuss our experiance with two such methods, Co-training and Active Learning. Co-training attempts to improve the proformance of learning algorithms when a small set of labeled and a large set of unlabeled instances are available. Active learning attempts to select a subset of instances from a large set of unlabeled instances to be labeled by a teacher. The goal is to achive the needed performance by significantlly less amount of labeled data comapred to regular supervised learning.